Yeah its an old one lol Glad you like it!
It didn't work on Pony so I had to wait for Illustrious to make it work in the XL world.

Thanks! I tried to pick a mix of popular characters, my favorites, and Pokemon -> because that's my main theme as a creator.

The approach you describe is the most reliable method. However for this one its trained entirely on realistic images.

It helps to give your images tags for things you want to control. For example, I tagged all the images in my dataset with "realistic, photo (medium), photo background" so by simply not using those tags in my prompt it leans towards not making those things.

Also my CAT Checkpoint tends to make things more anime. Its a pretty good checkpoint. (shameless self-advertising lol)

Finally, this lora is the 6th epoch. Normally my loras are the 10th epoch but the 10th one was too realistic so I picked a "less strong" epoch.
If you want you can check the training metadata of most loras (SD, SDXL, Pony, Illustrious) using this tool https://citronlegacy.github.io/lora-metadata-viewer/

If you drop my Lora into this tool you'll be able to see all the settings I used to train it and even the way I tagged my training images.

@javbukkake Oh! Keep using Xypher7's app. Mine is just a fork of that code with a small enhancement. (I have the credits at the bottom of my app linking to the original).

Good luck with your project! Feel free to come back and ask a question whenever. I'm always happy to help (whenever the notifications dont get lost lol)